Visualization Regularizers for Neural Network based Image Recognition
This work addresses the need for better regularization techniques in computer vision tasks, but it is incremental as it builds on known concepts like Tikhonov regularization.
The paper tackles the problem of improving classification accuracy in neural networks for image recognition by introducing a visualization regularizer that enforces smoothness in learned features, achieving higher accuracy compared to existing regularizers like L2 norm and dropout on benchmark datasets without increasing computational complexity.
The success of deep neural networks is mostly due their ability to learn meaningful features from the data. Features learned in the hidden layers of deep neural networks trained in computer vision tasks have been shown to be similar to mid-level vision features. We leverage this fact in this work and propose the visualization regularizer for image tasks. The proposed regularization technique enforces smoothness of the features learned by hidden nodes and turns out to be a special case of Tikhonov regularization. We achieve higher classification accuracy as compared to existing regularizers such as the L2 norm regularizer and dropout, on benchmark datasets without changing the training computational complexity.